Missing data were replaced with baseline data in the intention-to-treat analysis.
For the other items (poultry meat and fish) missing information of price data were replaced using the approach presented in Majumder et al. (2012) who extended the approach of Cox and Wohlgenant (1986).
In order to avoid duplicates in the analysis, probes representing the same gene symbol in Illumina™ data were replaced with their average intensity before applying the GSEA.
Analysis of repeated measures was on an intention-to-treat basis, in which missing data were replaced by the last value carried forward [ 50].
In all, missing data were replaced on average for 0.3% of the observation period, and 103 (3%) participants had missing data replaced for 1% to 5% of the period and 22 (0.6%) for 5% to 27% of the period.
Cox regression was performed on four datasets: the dataset with complete data, the dataset in which missing data were replaced by the mean (MS) and the multiply-imputed datasets – those imputed with and without including survival time in the imputation model (MI+ and MI−).
In all, missing data were replaced for 1 2 hours for 26.2% of the participants, >2 5 hours for 1.4% of the participants, >5 10 hours for 1.1% of the participants, and >10 25 hours for 0.4% of the participants.
Changes in weight and levels of engagement with the Internet-based treatment system were examined via separate repeated-measures ANOVA that followed the intent-to-treat (ITT) principle in which missing data were replaced by baseline weights (assuming no weight loss).
In these analyses, missing data were replaced by observed data, using maximum likelihood estimations (Twisk, 2003).
Less than 0.2% of data were replaced.
